Ethiopian Finger Spelling Classification: A Study to Automate Ethiopian Sign Language
No Thumbnail Available
Date
2008-09
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Ethiopian Finger Spelling is one of the communication means used among Ethiopian deaf. It is used to express names and any concepts that do not have sign in Ethiopian Sign Language. To fill the communication gap that exists among the deaf and between the deaf and the hearing, the Ethiopian Finger Spelling is processed using techniques from image processing and pattern processing. In this thesis work, a new attempt is done to design the architecture and select the appropriate techniques for each component of the Ethiopian Finger spelling classification system. The proposed architecture has components for image capturing, feature extraction, hand detection, region of interest segmentation and sign classification. For the tasks of hand detection and sign classification, experiments are conducted to select the appropriate pattern classifier and feature. In addition, the capability of the principal component analysis (PCA) driven and harr-like feature with neural network is tested through experiment.
According to the experimental result, neural network pattern processing techniques have high detection and classification rate when compared to the cascaded boosted classifier and template matching techniques for the task of hand detection and sign classification respectively. The overall hand detection rate of 99.43%, 96.59% and 77.27% were obtained using neural network with PCA driven feature, neural network with harr-like feature and boosted cascaded classifier respectively. In addition to this, the overall sign classification rate of 88.08%, 96.22% and 51.44% were obtained using neural network with PCA driven feature, neural network with harr-like feature and template matching respectively. In particular, neural network that use harr-like feature shows better performance for the task of sign classification and neural network with PCA driven feature shows better performance for the task of hand detection.
viii
Keywords: Ethiopian Sign Language, Ethiopian Finger Spelling, Hand detection and Sign Classification
Description
Keywords
Ethiopian Sign Language; Ethiopian Finger Spelling; Hand Detection and Sign Classification